opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/SnapshotManager.java

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8
   9 package org.opendaylight.controller.cluster.raft;
  10
  11 import akka.persistence.SnapshotSelectionCriteria;
  12 import com.google.common.annotations.VisibleForTesting;
  13 import java.util.List;
  14 import java.util.function.Consumer;
  15 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  16 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  17 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  18 import org.opendaylight.controller.cluster.raft.base.messages.SnapshotComplete;
  19 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  20 import org.slf4j.Logger;
  21
  22 public class SnapshotManager implements SnapshotState {
  23
  24     private final SnapshotState IDLE = new Idle();
  25     private final SnapshotState PERSISTING = new Persisting();
  26     private final SnapshotState CREATING = new Creating();
  27
  28     private final Logger LOG;
  29     private final RaftActorContext context;
  30     private final LastAppliedTermInformationReader lastAppliedTermInformationReader =
  31             new LastAppliedTermInformationReader();
  32     private final ReplicatedToAllTermInformationReader replicatedToAllTermInformationReader =
  33             new ReplicatedToAllTermInformationReader();
  34
  35
  36     private SnapshotState currentState = IDLE;
  37     private CaptureSnapshot captureSnapshot;
  38     private long lastSequenceNumber = -1;
  39
  40     private Runnable createSnapshotProcedure;
  41
  42     private ApplySnapshot applySnapshot;
  43     private Consumer<byte[]> applySnapshotProcedure;
  44
  45     public SnapshotManager(RaftActorContext context, Logger logger) {
  46         this.context = context;
  47         this.LOG = logger;
  48     }
  49
  50     public boolean isApplying() {
  51         return applySnapshot != null;
  52     }
  53
  54     @Override
  55     public boolean isCapturing() {
  56         return currentState.isCapturing();
  57     }
  58
  59     @Override
  60     public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
  61         return currentState.captureToInstall(lastLogEntry, replicatedToAllIndex, targetFollower);
  62     }
  63
  64     @Override
  65     public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
  66         return currentState.capture(lastLogEntry, replicatedToAllIndex);
  67     }
  68
  69     @Override
  70     public void apply(ApplySnapshot snapshot) {
  71         currentState.apply(snapshot);
  72     }
  73
  74     @Override
  75     public void persist(final byte[] snapshotBytes, final long totalMemory) {
  76         currentState.persist(snapshotBytes, totalMemory);
  77     }
  78
  79     @Override
  80     public void commit(final long sequenceNumber) {
  81         currentState.commit(sequenceNumber);
  82     }
  83
  84     @Override
  85     public void rollback() {
  86         currentState.rollback();
  87     }
  88
  89     @Override
  90     public long trimLog(final long desiredTrimIndex) {
  91         return currentState.trimLog(desiredTrimIndex);
  92     }
  93
  94     public void setCreateSnapshotRunnable(Runnable createSnapshotProcedure) {
  95         this.createSnapshotProcedure = createSnapshotProcedure;
  96     }
  97
  98     public void setApplySnapshotConsumer(Consumer<byte[]> applySnapshotProcedure) {
  99         this.applySnapshotProcedure = applySnapshotProcedure;
 100     }
 101
 102     public long getLastSequenceNumber() {
 103         return lastSequenceNumber;
 104     }
 105
 106     @VisibleForTesting
 107     public CaptureSnapshot getCaptureSnapshot() {
 108         return captureSnapshot;
 109     }
 110
 111     private boolean hasFollowers(){
 112         return context.hasFollowers();
 113     }
 114
 115     private String persistenceId(){
 116         return context.getId();
 117     }
 118
 119     public CaptureSnapshot newCaptureSnapshot(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex,
 120             boolean installSnapshotInitiated) {
 121         TermInformationReader lastAppliedTermInfoReader =
 122                 lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
 123                         lastLogEntry, hasFollowers());
 124
 125         long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
 126         long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
 127
 128         TermInformationReader replicatedToAllTermInfoReader =
 129                 replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
 130
 131         long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
 132         long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
 133
 134         List<ReplicatedLogEntry> unAppliedEntries = context.getReplicatedLog().getFrom(lastAppliedIndex + 1);
 135
 136         long lastLogEntryIndex = lastAppliedIndex;
 137         long lastLogEntryTerm = lastAppliedTerm;
 138         if(lastLogEntry != null) {
 139             lastLogEntryIndex = lastLogEntry.getIndex();
 140             lastLogEntryTerm = lastLogEntry.getTerm();
 141         } else {
 142             LOG.debug("{}: Capturing Snapshot : lastLogEntry is null. Using lastAppliedIndex {} and lastAppliedTerm {} instead.",
 143                     persistenceId(), lastAppliedIndex, lastAppliedTerm);
 144         }
 145
 146         return new CaptureSnapshot(lastLogEntryIndex, lastLogEntryTerm, lastAppliedIndex, lastAppliedTerm,
 147                 newReplicatedToAllIndex, newReplicatedToAllTerm, unAppliedEntries, installSnapshotInitiated);
 148     }
 149
 150     private class AbstractSnapshotState implements SnapshotState {
 151
 152         @Override
 153         public boolean isCapturing() {
 154             return true;
 155         }
 156
 157         @Override
 158         public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
 159             LOG.debug("capture should not be called in state {}", this);
 160             return false;
 161         }
 162
 163         @Override
 164         public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
 165             LOG.debug("captureToInstall should not be called in state {}", this);
 166             return false;
 167         }
 168
 169         @Override
 170         public void apply(ApplySnapshot snapshot) {
 171             LOG.debug("apply should not be called in state {}", this);
 172         }
 173
 174         @Override
 175         public void persist(final byte[] snapshotBytes, final long totalMemory) {
 176             LOG.debug("persist should not be called in state {}", this);
 177         }
 178
 179         @Override
 180         public void commit(final long sequenceNumber) {
 181             LOG.debug("commit should not be called in state {}", this);
 182         }
 183
 184         @Override
 185         public void rollback() {
 186             LOG.debug("rollback should not be called in state {}", this);
 187         }
 188
 189         @Override
 190         public long trimLog(final long desiredTrimIndex) {
 191             LOG.debug("trimLog should not be called in state {}", this);
 192             return -1;
 193         }
 194
 195         protected long doTrimLog(final long desiredTrimIndex) {
 196             //  we would want to keep the lastApplied as its used while capturing snapshots
 197             long lastApplied = context.getLastApplied();
 198             long tempMin = Math.min(desiredTrimIndex, (lastApplied > -1 ? lastApplied - 1 : -1));
 199
 200             if(LOG.isTraceEnabled()) {
 201                 LOG.trace("{}: performSnapshotWithoutCapture: desiredTrimIndex: {}, lastApplied: {}, tempMin: {}",
 202                         persistenceId(), desiredTrimIndex, lastApplied, tempMin);
 203             }
 204
 205             if (tempMin > -1 && context.getReplicatedLog().isPresent(tempMin)) {
 206                 LOG.debug("{}: fakeSnapshot purging log to {} for term {}", persistenceId(), tempMin,
 207                         context.getTermInformation().getCurrentTerm());
 208
 209                 //use the term of the temp-min, since we check for isPresent, entry will not be null
 210                 ReplicatedLogEntry entry = context.getReplicatedLog().get(tempMin);
 211                 context.getReplicatedLog().snapshotPreCommit(tempMin, entry.getTerm());
 212                 context.getReplicatedLog().snapshotCommit();
 213                 return tempMin;
 214             }
 215
 216             final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 217             if(tempMin > currentBehavior.getReplicatedToAllIndex()) {
 218                 // It's possible a follower was lagging and an install snapshot advanced its match index past
 219                 // the current replicatedToAllIndex. Since the follower is now caught up we should advance the
 220                 // replicatedToAllIndex (to tempMin). The fact that tempMin wasn't found in the log is likely
 221                 // due to a previous snapshot triggered by the memory threshold exceeded, in that case we
 222                 // trim the log to the last applied index even if previous entries weren't replicated to all followers.
 223                 currentBehavior.setReplicatedToAllIndex(tempMin);
 224             }
 225             return -1;
 226         }
 227     }
 228
 229     private class Idle extends AbstractSnapshotState {
 230
 231         @Override
 232         public boolean isCapturing() {
 233             return false;
 234         }
 235
 236         private boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
 237             captureSnapshot = newCaptureSnapshot(lastLogEntry, replicatedToAllIndex, targetFollower != null);
 238
 239             if(captureSnapshot.isInstallSnapshotInitiated()) {
 240                 LOG.info("{}: Initiating snapshot capture {} to install on {}",
 241                         persistenceId(), captureSnapshot, targetFollower);
 242             } else {
 243                 LOG.info("{}: Initiating snapshot capture {}", persistenceId(), captureSnapshot);
 244             }
 245
 246             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 247
 248             LOG.debug("{}: lastSequenceNumber prior to capture: {}", persistenceId(), lastSequenceNumber);
 249
 250             SnapshotManager.this.currentState = CREATING;
 251
 252             try {
 253                 createSnapshotProcedure.run();
 254             } catch (Exception e) {
 255                 SnapshotManager.this.currentState = IDLE;
 256                 LOG.error("Error creating snapshot", e);
 257                 return false;
 258             }
 259
 260             return true;
 261         }
 262
 263         @Override
 264         public boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex) {
 265             return capture(lastLogEntry, replicatedToAllIndex, null);
 266         }
 267
 268         @Override
 269         public boolean captureToInstall(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
 270             return capture(lastLogEntry, replicatedToAllIndex, targetFollower);
 271         }
 272
 273         @Override
 274         public void apply(ApplySnapshot applySnapshot) {
 275             SnapshotManager.this.applySnapshot = applySnapshot;
 276
 277             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 278
 279             LOG.debug("lastSequenceNumber prior to persisting applied snapshot: {}", lastSequenceNumber);
 280
 281             context.getPersistenceProvider().saveSnapshot(applySnapshot.getSnapshot());
 282
 283             SnapshotManager.this.currentState = PERSISTING;
 284         }
 285
 286         @Override
 287         public String toString() {
 288             return "Idle";
 289         }
 290
 291         @Override
 292         public long trimLog(final long desiredTrimIndex) {
 293             return doTrimLog(desiredTrimIndex);
 294         }
 295     }
 296
 297     private class Creating extends AbstractSnapshotState {
 298
 299         @Override
 300         public void persist(final byte[] snapshotBytes, final long totalMemory) {
 301             // create a snapshot object from the state provided and save it
 302             // when snapshot is saved async, SaveSnapshotSuccess is raised.
 303
 304             Snapshot snapshot = Snapshot.create(snapshotBytes,
 305                     captureSnapshot.getUnAppliedEntries(),
 306                     captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
 307                     captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm(),
 308                     context.getTermInformation().getCurrentTerm(),
 309                     context.getTermInformation().getVotedFor(), context.getPeerServerInfo(true));
 310
 311             context.getPersistenceProvider().saveSnapshot(snapshot);
 312
 313             LOG.info("{}: Persisting of snapshot done: {}", persistenceId(), snapshot);
 314
 315             long dataThreshold = totalMemory *
 316                     context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 317             boolean dataSizeThresholdExceeded = context.getReplicatedLog().dataSize() > dataThreshold;
 318
 319             boolean logSizeExceededSnapshotBatchCount =
 320                     context.getReplicatedLog().size() >= context.getConfigParams().getSnapshotBatchCount();
 321
 322             final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 323             if (dataSizeThresholdExceeded || logSizeExceededSnapshotBatchCount) {
 324                 if(LOG.isDebugEnabled()) {
 325                     if(dataSizeThresholdExceeded) {
 326                         LOG.debug("{}: log data size {} exceeds the memory threshold {} - doing snapshotPreCommit with index {}",
 327                                 context.getId(), context.getReplicatedLog().dataSize(), dataThreshold,
 328                                 captureSnapshot.getLastAppliedIndex());
 329                     } else {
 330                         LOG.debug("{}: log size {} exceeds the snapshot batch count {} - doing snapshotPreCommit with index {}",
 331                                 context.getId(), context.getReplicatedLog().size(),
 332                                 context.getConfigParams().getSnapshotBatchCount(), captureSnapshot.getLastAppliedIndex());
 333                     }
 334                 }
 335
 336                 // We either exceeded the memory threshold or the log size exceeded the snapshot batch
 337                 // count so, to keep the log memory footprint in check, clear the log based on lastApplied.
 338                 // This could/should only happen if one of the followers is down as normally we keep
 339                 // removing from the log as entries are replicated to all.
 340                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
 341                         captureSnapshot.getLastAppliedTerm());
 342
 343                 // Don't reset replicatedToAllIndex to -1 as this may prevent us from trimming the log after an
 344                 // install snapshot to a follower.
 345                 if(captureSnapshot.getReplicatedToAllIndex() >= 0) {
 346                     currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 347                 }
 348
 349             } else if(captureSnapshot.getReplicatedToAllIndex() != -1){
 350                 // clear the log based on replicatedToAllIndex
 351                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(),
 352                         captureSnapshot.getReplicatedToAllTerm());
 353
 354                 currentBehavior.setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 355             } else {
 356                 // The replicatedToAllIndex was not found in the log
 357                 // This means that replicatedToAllIndex never moved beyond -1 or that it is already in the snapshot.
 358                 // In this scenario we may need to save the snapshot to the akka persistence
 359                 // snapshot for recovery but we do not need to do the replicated log trimming.
 360                 context.getReplicatedLog().snapshotPreCommit(context.getReplicatedLog().getSnapshotIndex(),
 361                         context.getReplicatedLog().getSnapshotTerm());
 362             }
 363
 364             LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex: {} " +
 365                     "and term: {}", context.getId(), context.getReplicatedLog().getSnapshotIndex(),
 366                     context.getReplicatedLog().getSnapshotTerm());
 367
 368             if (context.getId().equals(currentBehavior.getLeaderId())
 369                     && captureSnapshot.isInstallSnapshotInitiated()) {
 370                 // this would be call straight to the leader and won't initiate in serialization
 371                 currentBehavior.handleMessage(context.getActor(), new SendInstallSnapshot(snapshot));
 372             }
 373
 374             captureSnapshot = null;
 375             SnapshotManager.this.currentState = PERSISTING;
 376         }
 377
 378         @Override
 379         public String toString() {
 380             return "Creating";
 381         }
 382
 383     }
 384
 385     private class Persisting extends AbstractSnapshotState {
 386
 387         @Override
 388         public void commit(final long sequenceNumber) {
 389             LOG.debug("{}: Snapshot success -  sequence number: {}", persistenceId(), sequenceNumber);
 390
 391             if(applySnapshot != null) {
 392                 try {
 393                     Snapshot snapshot = applySnapshot.getSnapshot();
 394
 395                     //clears the followers log, sets the snapshot index to ensure adjusted-index works
 396                     context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context));
 397                     context.setLastApplied(snapshot.getLastAppliedIndex());
 398                     context.setCommitIndex(snapshot.getLastAppliedIndex());
 399                     context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
 400
 401                     if(snapshot.getServerConfiguration() != null) {
 402                         context.updatePeerIds(snapshot.getServerConfiguration());
 403                     }
 404
 405                     if(snapshot.getState().length > 0 ) {
 406                         applySnapshotProcedure.accept(snapshot.getState());
 407                     }
 408
 409                     applySnapshot.getCallback().onSuccess();
 410                 } catch (Exception e) {
 411                     LOG.error("{}: Error applying snapshot", context.getId(), e);
 412                 }
 413             } else {
 414                 context.getReplicatedLog().snapshotCommit();
 415             }
 416
 417             context.getPersistenceProvider().deleteSnapshots(new SnapshotSelectionCriteria(
 418                     sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), Long.MAX_VALUE, 0L, 0L));
 419
 420             context.getPersistenceProvider().deleteMessages(lastSequenceNumber);
 421
 422             snapshotComplete();
 423         }
 424
 425         @Override
 426         public void rollback() {
 427             // Nothing to rollback if we're applying a snapshot from the leader.
 428             if(applySnapshot == null) {
 429                 context.getReplicatedLog().snapshotRollback();
 430
 431                 LOG.info("{}: Replicated Log rolled back. Snapshot will be attempted in the next cycle." +
 432                         "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
 433                         context.getReplicatedLog().getSnapshotIndex(),
 434                         context.getReplicatedLog().getSnapshotTerm(),
 435                         context.getReplicatedLog().size());
 436             } else {
 437                 applySnapshot.getCallback().onFailure();
 438             }
 439
 440             snapshotComplete();
 441         }
 442
 443         private void snapshotComplete() {
 444             lastSequenceNumber = -1;
 445             applySnapshot = null;
 446             SnapshotManager.this.currentState = IDLE;
 447
 448             context.getActor().tell(SnapshotComplete.INSTANCE, context.getActor());
 449         }
 450
 451         @Override
 452         public String toString() {
 453             return "Persisting";
 454         }
 455
 456     }
 457
 458     private static interface TermInformationReader {
 459         long getIndex();
 460         long getTerm();
 461     }
 462
 463     static class LastAppliedTermInformationReader implements TermInformationReader{
 464         private long index;
 465         private long term;
 466
 467         public LastAppliedTermInformationReader init(ReplicatedLog log, long originalIndex,
 468                                          ReplicatedLogEntry lastLogEntry, boolean hasFollowers){
 469             ReplicatedLogEntry entry = log.get(originalIndex);
 470             this.index = -1L;
 471             this.term = -1L;
 472             if (!hasFollowers) {
 473                 if(lastLogEntry != null) {
 474                     // since we have persisted the last-log-entry to persistent journal before the capture,
 475                     // we would want to snapshot from this entry.
 476                     index = lastLogEntry.getIndex();
 477                     term = lastLogEntry.getTerm();
 478                 }
 479             } else if (entry != null) {
 480                 index = entry.getIndex();
 481                 term = entry.getTerm();
 482             } else if(log.getSnapshotIndex() > -1){
 483                 index = log.getSnapshotIndex();
 484                 term = log.getSnapshotTerm();
 485             }
 486             return this;
 487         }
 488
 489         @Override
 490         public long getIndex(){
 491             return this.index;
 492         }
 493
 494         @Override
 495         public long getTerm(){
 496             return this.term;
 497         }
 498     }
 499
 500     private static class ReplicatedToAllTermInformationReader implements TermInformationReader{
 501         private long index;
 502         private long term;
 503
 504         ReplicatedToAllTermInformationReader init(ReplicatedLog log, long originalIndex){
 505             ReplicatedLogEntry entry = log.get(originalIndex);
 506             this.index = -1L;
 507             this.term = -1L;
 508
 509             if (entry != null) {
 510                 index = entry.getIndex();
 511                 term = entry.getTerm();
 512             }
 513
 514             return this;
 515         }
 516
 517         @Override
 518         public long getIndex(){
 519             return this.index;
 520         }
 521
 522         @Override
 523         public long getTerm(){
 524             return this.term;
 525         }
 526     }
 527 }